Agents

Rescriptum's agent system is the core of the IDE. Unlike simple chat-based code assistants, agents in Rescriptum are autonomous workflows that can plan, execute, test, and iterate across your entire codebase.

How agents work

When you send a message, the agent enters a loop: it reads files, writes code, runs commands, and evaluates results. Each step is visible in the chat as an activity card showing what tools were used and what happened.

Tool system

Agents have access to a set of tools: read_file, write_file, run_command, search_codebase, find_symbols, and more. Each tool call can require approval, be auto-approved, or earn trust over time through the journal system.

Plan mode

Toggle Plan mode to have the agent generate a structured plan before executing. The plan is saved as plan.md in your project root and the agent follows it step by step.

Long-horizon tasks

Agents can handle multi-step tasks that span many files and commands. The hung-task recovery system monitors progress and retries or terminates stalled operations automatically.